Generative Computing Lab
At the Generative Computing Lab (GCL), we build human-centric solutions to make Generative AI more accessible and impactful for both the public and expert professionals. Above all, we prioritize the responsible use of Generative AI, developing robust frameworks to safeguard data security and ensure ethical deployment.
Research Highlights
StyleComposer: Training-Free Multi-Reference Style Composition (arXiv preprint:2608.05213)Enabling fine-grained style composition by disentangling and combining visual attributes from multiple references without additional training.
Compositional Image Synthesis with Inference-Time Scaling (ICASSP 2026)Enhancing text-to-image compositionality via iterative self-refinement and inference-time scaling without additional training.
Nearly Zero-Cost Protection Against Mimicry by Personalized Diffusion Models (CVPR 2025)An ultra-fast method to protect images against mimicry attempts from personalized diffusion models.
DreamStyler: Paint by Style Inversion with Text-to-Image Diffusion Models (AAAI 2024)An innovative approach to image generation that leverages textual inversion within text-to-image diffusion models.
DiffBlender: Scalable and Composable Multimodal Text-to-Image Diffusion Models (Expert Systems with Applications)A novel framework for blending multiple modalities in text-to-image diffusion models.
AesPA-Net: Aesthetic Pattern-Aware Style Transfer Networks (ICCV 2023)Aesthetic pattern-aware style transfer networks for high-quality image synthesis.
Interactive Cartoonization with Controllable Perceptual Factors (CVPR 2023)An interactive way of cartoonization that allows for the manipulation of perceptual factors.
WebtoonMe: A Data-Centric Approach for Full-Body Portrait Stylization (SIGGRAPH Asia TC 2022)A practical approach to full-body portrait stylization.
Rethinking Data Augmentation for Image Super-Resolution: A Comprehensive Analysis and a New Strategy (CVPR 2020)GCL News
May 2026
A paper Compositional Image Synthesis with Inference-Time Scaling has been accepted to ICASSP 2026.
Mar 2026
A paper T2LF: LLM-Guided Multimodal Diffusion for Text-to-Light Field Synthesis has been selected as a Best Paper Award Finalist at WACV 2026 (13/832 β 1.6%).
Feb 2026
Professor Ahn has been invited to give a talk titled "Securing Visual Copyright: Recent Advances in Image Protection Against Diffusion Models" at IPIU 2026.
Oct 2025
Professor Ahn has been invited to give a tutorial titled "Responsible Generative AI: From Provenance to Protection" at ICCE-Asia 2025.
Aug 2025
A paper DiffBlender: Composable and Versatile Multimodal Text-to-Image Diffusion Models has been accepted to Expert Systems with Applications (ESWA).